community or connections? a social networks approach to chain migration

8/8/2019 Community or Connections? A Social Networks Approach to Chain Migration

1/47

Community or Connections?

A Social Networks Approach to

Chain Migration

Randall S. Kuhn

Population Program Institute of Behavioral Science

University of Colorado

484 UCBBoulder, CO 80309-0484

[email protected]

September 2008


2/47

Abstract

This paper introduces a social networks approach to specifying and measuring the process of

chain migration. Using a unique combination of survey and vital registration data for the Matlab

Subdistrict of Bangladesh, I construct measures of past migrant stock that are specific to

household lineage and residential compound groups, and to social networks that share historical

links to those groups. Regression models predict the hazard of subsequent urban-rural and

international migration in terms of migrant stock for these newly specified networks alongside

effects for the village and household. Migrant stock for residential compound and social

networks are strong predictors of the hazard of migration, particularly of rural-urban migration,

explaining away about 50% of the village migrant stock effect. The demonstrate the potential

value of social network-based measures of social capital, the extraordinary benefits of using

Demographic Surveillance System data for migration research, and concerns regarding a village-

based specification of migration-specific social capital.


3/47

I. Introduction

Some of the most compelling results to emerge from recent migration research have come in

explaining chain migration, the tendency for past migration to be strongly associated with further

migration. Although we can observe this process unfolding as people from a single nation,

region, and even community come to dominate certain industries in migrant destination areas,

only recently have we developed detailed testable, theoretical hypotheses regarding chain

migration.

Chain migration unfolds as the costs and risks of migration decline or the returns to

migration increase for members of a contextual unit, such as a community. Among other things,

we have learned to conceive of migration as a social diffusion process whereby low status

households initially not migrating will eventually gain access to migration and all of its benefits

(Durand, Massey, and Zenteno 2001). Information and opportunities relating to migration

become institutionalized as a form of migration-specific social capital potentially benefiting

the entire community (Massey, Goldring, and Durand 1994; Massey et al. 1987). The penetration

of migration opportunities throughout the socioeconomic distribution of a community or society

is one critical element for anticipating migrations ultimate impact on socioeconomic

stratification and its impact on the welfare of the poorest households. While this perspective is

particularly constructive for seeing how receiving societies interpret the growth of migration

from a single sending society, region, or even community, it may offer only a partial

understanding of the process form the sending community perspective.

Treating chain migration as a process of transmission of contextual social capital

overlooks the possibility that migration opportunities are not common assets available to all who

wish to use them, but scarce commodities. While migration opportunities may not be sold on the


4/47

2

open market, they may be allocated with extreme preference: to members of a community with a

shared lineal or social history, and to members of a community with resources to offer in explicit

or implicit exchange. Furthermore, some members of a community with a limited migration

experience may have substantial ties to past migrants from other communities. Extra-community

ties, not captured by contextual measures of migrant stock, help explain how migration

opportunities are transmitted from community to community. These two concerns are not

unrelated: if low status households in a high-migration community do not provide a suitable

match for a migration-specific social network, these opportunities may instead pass to high status

households in low-migration communities.

This paper serves as a first step in beginning to address the extent to which the process of

chain migration reflects a contextual or a conditional process. Taking advantage of the unique

historical demographic detail of the Matlab Health and Demographic Surveillance System

(MHDSS), I predict the hazard of rural-urban and international migration using contextual

measures of migrant stock as well as network-based measures. Household are linked in primary

lineage networks according to a shared history of coresidence in family and compound units.

Primary lineage networks are then linked to one another through histories of inter-marriage and

internal migration to generatepeer networks.

The following two sections present potential explanations for chain migration in greater

detail and draw on previous qualitative research to stress the importance of chain migration in

the context of rural Bangladesh. The fourth section describes the survey and surveillance data

used in the study and offers extensive detail on the construction of peer networks and the

estimation of migrant stock. Finally, contextual and peer networks data are used to predict the


5/47

3

hazard of rural-urban and international migration among a sample of adults males in Matlab from

1996 to 2002.

II. Contextual Relationships between Past and Future Migration

Past research has observed the association between higher levels of past migration and higher

levels of subsequent migration at the household, community, regional, and national level

(Massey 1990; Massey and Espinosa 1997; Wegge 1998; Massey et al. 1999). Studies have

typically predicted the likelihood of migration in terms of the accumulated migration experience

(migrant stock) among family units and among contextual units such as communities (Munshi

2003; Zhao 2003). The proximate effects leading from past migration to subsequent migration

include lowering moving costs (Carrington, Detragiache, and Vishwanath 1996; Wegge 1998;

Shah and Menon 1999); increasing the chances of permanent employment and expected wages

(Taylor 1986; Greenwell, Valdez, and DaVanzo 1997; Shah and Menon 2000; Sanders, Nee, and

Sernau 2002; Aguilera and Massey 2003; Aguilera 2003); and enhancing socialization and

adaptation in the destination area (Sanders et al. 2002; Menjivar 2002).

Both structural and social mechanisms can explain these proximate effects. Structural

mechanisms underlying chain migration reflect accumulated changes to local markets,

infrastructure, culture, or knowledge that would permanently change the calculus of the

migration decision (Massey 1990).1

For example, high demand for transport between origin and

destination areas may lead to improvements in transport infrastructure that reduce the costs of

moving for future migrants. As individual changes accumulate, migration may engender a

culture of migration whereby life is increasingly oriented around the eventual practice of

1 For a more detailed treatment of structural factors underlying chain migration, see Massey (1990).


6/47

4

migration (Kandel and Massey 2002). Empirical models have captured the structural effects of

migration either by controlling for contextual measures of migrant stock or by controlling for

specific changes to the context (Lindstrom and Lauster 2001; Shah and Menon 1999).

The social effects underlying chain migration stem from the contacts that members of a

sending community maintain with current and returned migrants (Espinosa and Massey 1997;

Lindstrom and Lauster 2001; Shah and Menon 2000). In this respect, chain migration bears

characteristics of a process of social diffusion (Massey et al. 1987; Van der Gaag and Van

Wissen 2002). Early movers tend to come from the highest strata of a community, as household

wealth may buffer prospective migrants against the costs and risks associated with moving

(Taylor 1986; Waldorf 1996; Massey et al. 1999; Sanders et al. 2002). Over time, social

networks should reduce the costs and risks of migration for an increasing share of the population,

and to an increasingly representative cross-section of the community (Winters, de Janvry, and

Sadoulet 2001; Zhao 2003).

As the number of individuals in a local area network having knowledge, experience, or

authority in a specific destination area increases and reaches a critical density, individual social

connections can be converted into a form of migration-specific social capital based on shared

community or ethnic origins (Ebaugh 2000; Winters et al. 2001; Elmhirst 2002; Davis, Stecklov,

and Winters 2002; Hagan and Ebaugh 2003). For example, Kuhns (2003) qualitative analysis of

migration in Bangladesh demonstrates how community-based identities that had no currency in a

strictly-local context can serve as entrees into migration and employment opportunities as the

density of people from the same rural Bangladeshi communities living in Dhaka, the capital,

increased. This and other research provide a compelling reason to treat migration-specific social


7/47

5

capital as a contextual characteristic, to be measured in the same fashion as structural effects

(Wilson 1998).

Yet more recently, considerable concern has been expressed over simplifying social

capital to a contextual concept whereby all members of the community draw equal benefit

(Portes 1997, 2000; Krissman 2005). Ignoring, for the moment, the possibility that social capital

can have negative implications, concerns have been raised about the presence of social exclusion

and conditionality in social networks, particularly those that ascribe a particular and potentially

lucrative opportunity such as migration (Burt 1998, 2000; Wejnert 2002; Krishna 2004).

Contextual measures of social capital may ignore the explicit exclusion of certain subgroups in a

community from the channels of social capital (Winters et al. 2002; Krishna 2004; Das 2004).

Some have sought to address the role of exclusion in networks of socially-transmitted

opportunity by explicitly disaggregating networks according to observed and acknowledge social

cleavages (Curran and Rivero-Fuentes 2003; Davis and Winters 2001; Silvey and Elmhirst

2003). This approach has proven quite effective for separating communities along discrete, easily

differentiated groupings such as gender and ethnicity. It cannot address the unobserved or

continuous patterns of social exclusion characterizing social lineage, social class, and social peer

relationships in rural communities. Some members of a community will have much larger social

networks, and some networks will have a higher proportion of experienced individuals (Portes

1998; Strang and Soule 1998; Winters et al. 2001). Some of this variation should be random, but

some of it should be systematic. If high status households move first, then those with ties to high

status households will also have stronger ties to past migrants. The size, status, and migration

experience of a social network should be correlated positively with a constellation of other


8/47

6

measures of socioeconomic status. Contextual measures of migration-specific social capital offer

no potential for households in a community to vary on any of these social network dimensions.

A few papers have addressed the importance of weak and strong ties in chain migration,

but most have focused on differences between family networks and community networks that are

specified contextually (Wilson 1998; Winters et al. 2001; Collyer 2005). Yet one critical factor

in the transmission of migration opportunities between communities may be the strength of

connection to social peers living outside the community of origin. Among communities with low

levels of migration-specific social capital, contextual measures of migrant stock offer no

potential to discriminate among households in a community having strong migration-specific

social ties outside the community, and those that dont. Yet our accumulated knowledge on the

strength of weak ties suggests it is far more likely for the earliest migrants in a community to be

influenced by past migrants from other, more experienced communities than for them to move

without any social impetus at all.

III. Research Context

Migration to cities and other countries has become central to the economy of rural areas of

Bangladesh such as the Matlab Upazila study area.2

Matlab is a rural subdistrict located about 55

kilometers southeast of the capital city, Dhaka. Like most districts of rural Bangladesh, the

traditional basis of economic life is a single monsoon-fed rice crop although recent water

management projects have created opportunity for more extensive cropping in some areas. The

monsoon and persistent limitations to production technology and liquidity management leave

2Upazila, the lowest level of government, is comparable in size to a US county. It is also referred to as thana, which

means literally police stand.


9/47

7

households exposed to extreme seasonality in commodity supply, prices, and nutrition, as well as

persistent risk of economic failure.

While some of these deficiencies can be addressed by seasonal migration to richer rice-

producing regions of the country, a far more common and lucrative remedy is migration to an

urban or overseas destination. Although distance to the capital is short, travel time is six hours by

boat or bus, depending on the season, necessitating rural-urban migration rather than commuting.

Close proximity does facilitate a pattern of frequent urban-rural circulation, solo migration by

adult males both single and married, and a constant flow of information and opportunity between

destination cities and origin villages. International migration is also quite common in the study

area, most frequently to India (primarily among the Hindu population, 12% of the sample) and to

oil-rich countries in the Persian Gulf, particularly the Kingdom of Saudi Arabia. International

migration was practiced almost exclusively by males.

Previous qualitative research on social networks and rural-urban migration from Matlab

has addressed the structure and function of the origin-area social networks critical to gaining

access to rural-urban migration opportunities. The study highlights the role not merely of

residence, but of specific ties accessed through a shared lineal history or functional economic

linkages to past migrants, particularly to powerful migrant gatekeepers, such as business

owners and civil servants, responsible for bringing large numbers of migrants to the city. This

work also highlights the importance of marital alliances in the transmission of migration

opportunities. Marriages, which are typically extra-local, provide a powerful social linkage

between communities that would necessitate migrant stock measures that are not village-based.

Marital ties transmit migration opportunities from high-migration communities to low-migration


10/47

8

ones, bypassing members of high-migration communities that may not be willing or able to

participate in chain migration networks.

IV. Study Data

The analysis of migration-specific peer networks in Matlab takes advantage of a unique

combination of survey and surveillance data. The Matlab Health and Demographic Surveillance

System (HDSS) has computerized demographic histories for approximately 40,000 households

in 149 villages since 1966. This study takes advantage of data spanning 1974 to 2002. The

system gathers dates and basic information on every birth, death, marriage, divorce, and

migration. Migration surveillance captures all moves of more than six months in length, with

separate files for moves outside and within the boundaries of the HDSS area.

Unique identifiers for individuals, households, baris (residential compounds), and

villages facilitate the construction of primary and peer networks for the analysis. Censuses

conducted in 1982 and 1993 facilitate tracking of household partition and the creation of a

household lineage identifier for use in lieu of a current household identifier. HDSS also provided

the sampling frame for the Matlab Health and Socioeconomic Survey (MHSS), which provides

the baseline sample for the current study. See the section on building the dataset below for

further detail on MHSS and units of analysis.

It may be helpful to the reader to keep in mind some of the advantages and disadvantages

of using HDSS data to study migration, and to consider the likelihood of recreating some of its

advantages in the context of survey data collection. HDSS data are valuable for facilitating

prospective modeling of vital events. Large sample size facilitates the modeling of rare events.

Relatively common events such as migration can be disaggregated into specific events like rural-

urban and international migration. Although migration is relatively rare in any given year, we


11/47

9

can still conduct stable hazard models of migration without sampling on the dependent variable,

as in the case of ethnosurveys, the typical approach to studying migration.

The advantages most salient to this paper stem from the accumulation of prospectively-

gathered historical background data on the study population. HDSS data provide precise

migration histories for all individuals in the study population up to the point they leave the study

area and upon their return. Furthermore, we can aggregate migration and other vital events from

individual level to higher levels such as the household, the residential compound, and the village.

Aggregate vital event data of this form mirror the contextual measures of migration-specific

social capital used in previous studies, but with several added advantages. Prospective data

collection gives greater precision on the timing and characteristics of vital events. It also

facilitates well-defined identification of contextual units, and frequent updating of changes to

membership in contextual units. Finally, HDSS data provide 100% coverage of all events

occurring in the contextual units with little possibility of bias.

The key benefit of HDSS data for this study is the opportunity to identify direct, non-

contextual linkages between individuals or households that are not explicit to the data

framework. Namely, this study will construct peer networks from histories of inter-marriage and

internal movement between households. For a number of reasons, this approach is highly

experimental. First, and most obviously, no study has ever constructed peer networks from

HDSS data. Second, prior peer network studies have identified linkages directly through

nominations; this study uses only the indirect links available on hand.

The use of inter-marriage and internal migration histories can be justified on grounds

other than convenience. Studies of migration in Bangladesh and elsewhere in the Indian

Subcontinent have demonstrated their role as both a result of and a foundation for strategic


12/47


13/47

11

V. Building the Dataset

Table 1 outlines the steps involved in creating the data for this study. The description emphasizes

a careful and transparent description of the steps necessary to construct peer networks and

measures of migrant stock for all units of analysis. Table 1 identifies these units. The analysis

looks at two units commonly employed in the chain migration literature: the survey household

and the village. It also identifies each respondent with two primary social networks, the

household lineage and the bari. The use of the term primary network implies both that HDSS

identifies these linkages directly and that they correspond more of less to social groupings of

recognized importance within Bangladeshi society. Finally, peer networks are constructed for the

household lineage and bari. In contrast with primary networks, peer networks neither correspond

to any established social order nor are they identified directly by the DSS database.

[INSERT TABLE 1 ABOUT HERE]

Step 1: Identification of baseline study sample

The baseline population consists of all adult males, age 15 to 49, residing in households included

in the Matlab Health and Socioeconomic Survey (MHSS), a survey conducted in Matlab in 1996

by ICDDR,B, RAND, Harvard School of Public Health, and the University of Pennsylvania. The

primary sample consists of a two-stage random sample of 4,364 households in the 37,500

household HDSS study area.3

Household rosters gathered basic demographic and socioeconomic

data on each household member. The household head responded to a detailed module on the

3 The first stage was a random selection of baris. The second stage selected one household from the bari at random,

and a second household purposively with priority to fathers, children, and brothers of the head of the first household.

If no such household was present the second household was chosen at random. All analyses incorporate samplingweights to account for the likelihood that any given household in the sample was interviewed. An HDSS census of

households and baris, conducted in 1996, enabled the construction of highly precise weights.


14/47

12

economic activities of the household and each member. An adult survey module gathered

detailed data on health, economic activities, and demographic history. The adult module was

only administered to a complex selection of household members, resulting in only 66% of males

age 15-49 being interviewed.4

To take advantage of as many cases as possible, this analysis

includes all males age 15-49 rather than just those who responded to the individual adult survey.

While this limits the availability of highly detailed control variables, the rosters and household

economic survey should provide ample statistical control for the purposes of this analysis.

Step 2: Identification of Primary Social/Administrate Units

The second step in the data exercise entails identifying the primary units of analysis for the

construction of migrant stock data. First, we look at the traditional units of analysis. MHSS

identifies each respondent with hissurvey household. As in many other studies, the household is

uniquely defined as a group of individuals that eat from the same cooking pot, even though they

may lodge in separate dwellings. HDSS data identify thestudy village, which corresponds to the

mouza, an administrative unit that may include several neighborhoods.5

HDSS identified 149

villages in 1974, of which 141 remained after substantial flooding and river erosion in the

1980s.

The first primary network unit drawn from the HDSS data is the household lineage.

While individual survey households typically include a single nuclear family, HDSS records of

household splits occurring between 1974 and 1996 can identify the household from which the

current MHSS household emerged as of 1974. Distinct MHSS households can thus be joined to a

4 Selection proceeded as follows: head and spouse of head, members age 50+ and spouses, and two additional

randomly-selected adults age 15-49.5 To identify MHSS respondents with their villages and with all subsequent HDSS identifiers and peer networks,

MHSS data were linked to the privacy-protected, secure version of MHSS.


15/47

13

smaller number of lineages that reflect long-term social ties. In most cases, household splits trace

the fission of patrilineal families as sons separate from fathers, and brothers from brothers.

Identifying households that were part of the same household in 1974 is therefore principally a

way of identifying cousins. Household that had not split since 1974 formed no linkages, and

stood alone as the only household in their lineage.6

HDSS tracks each households membership in one further social grouping, the bari.Baris

are residential compounds or courtyards that form the locus of household production and ritual.

Baris are a spatial grouping, similar to a residential block in an urban context. The social

function of the bari varies from district to district, but in Matlab they typically reflect a shared

lineage, such as descendants of a single grandfather or great-grandfather. HDSS initially

identified baris in order to address the possibility that the arbitrary definition of a household

could not fully capture patrilineal social organization in the study area. Evidence from the

MHSS, which was designed to address the meaning and function of the bari, suggests substantial

pooling of socioeconomic risk among households in the same bari (Foster 2005; Joshi and Sinha

2005). HDSS updated bari identifiers in in the 1993 census.

Step 3: Peer Network Construction

The next step in the process involves the construction of peer networks joining respondents to

household lineages and baris that have historic connections to their own. While few data sources

provide data on the explicit linkages among members of such a large population, HDSS

marriages and internal migration histories provide a proxy for peer linkages. In general, these

6 For the most part, household splits in rural Bangladesh involve the formation of a new household in close

proximity to the original household. If a household split and the new household moved to a distinct spatial area, thetwo households would not share a household lineage. Their relationship would, however, be picked up by the peer

network linkag


16/47

14

events draw a history of social linkage between household lineages and baris. More specifically,

marriages, which typically involve mates from separate baris or villages, often involve the

transmission of migration opportunities, particularly from villages with high migrant stock to

those with low stock (Kuhn 2003).

First-order peer linkages, or directly linkages between two units, are formed from a

combination of internal migration and marriage data because internal migration data are

available only as far back as 1982. For the June 1982- December 1996, peer linkages are formed

using internal migration data, which include all marriages as well as a number of less common

internal movements.

7

For June 1974 to June 1982, peer linkages result only from marriages.

A first-order peer linkage is assigned for each internal migration or marriage episode

linking two household lineages orbari, capturing the strength of association between two units.

Rare cases in which multiple individuals moved between units in the same episode resulted in

only one linkage. The current dataset does not account for the timing of the formation of the peer

linkage, either with respect to the absolute timing of their linkage or to the relative timing in

relation to the external migration episodes that contribute to the stock measures. These concerns

provide fertile ground for future studies. Linkages formed between two survey households from

the same household linage orbari are not counted. Primary networks already capture these

linkages, and doing so would merely link units to themselves.

Second-order peer linkages are formed by merging the list of first-order peers to itself,

creating a list of peers peers. Second-order networks reflect a nexus of shares ties that may be

critical to the relaying of information and opportunity. In particular, second-order peer network

measures should compete with contextual village-level measures as a possible indicator of the

7 These include household relocation, movement of elders from sibling to sibling, and movement of children to live

with relatives living in closer proximity to school


17/47

15

transmission of migration through cultural or identity-based pathways. In other words, do

identity-based or community-based migration opportunities reflect a pattern of solidarity among

those of shared community origin, or those who share loose social linkages?

Because linking episodes were used to draw first-order peer linkages, two units can share

any number of first- and second-order linkages. As with first-order linkages, you cannot be your

own peers peer. Second-order linkages that run from your unit, to another unit, and back to your

own unit, even to a different survey household than your own, are not counted since they are

captured by primary network linkages.

Table 2 summarizes network size and migrant stock for each reference group. As

suggested above, primary household linage networks contain a substantially more males than the

survey household itself. Also, primary bari networks are substantially larger than primary

household lineage networks. As the number of network linkages increases exponentially with

each higher order linkage, and the number of males increases proportionately, baripeer networks

grow increasingly larger. The mean size of a second-orderbaripeer network is substantially

larger than the mean size of a village.


There is considerable variability in primary and peer network size, particularly for

second-order networks. A substantial proportion of networks include no males, even among the

fairly large first-orderbari networks. Since peer linkages are constructed recursively, households

with no primary bari links also have no peer links.

Step 5: Construction of Migrant History Files

Migration history or migrant stock data come from HDSS out-migration records for the period

from June 1974 to December 1996. HDSS out-migration files record all movements out of the


18/47

16

study area of six months of greater. The migrant stock file include only moves made by men

aged 15 to 49 at the time of the move. Destination codes identify moves to urban destinations

within Bangladesh (rural-urban migration) and overseas destinations (international migration).

This study does not address moves to other destinations outside the HDSS area but inside

Bangladesh. The file also does not distinguish among specific internal or international

destinations. It may be helpful to note that Dhaka, the capital city, was the destination of 69% of

rural-urban migrants. Among international migrants, 65% moved to nations in the Middle East,

primarily Saudi Arabia. Since most overseas migrants arrange their trips through manpower

agencies and not through foreign consulates, there are reasons to think that migration to any

foreign country confers advantages in moving to any other foreign country.

The exploratory nature of this analysis necessitated a number of decisions aimed at

simplifying the definition of migrant stock. First, the file does not distinguish between moves in

or out of the region, or between current and returned migrants.8

Second, the migrant stock file

accounts for whether an individual ever migrated to an urban or overseas destination, not for the

total number of episodes to each destination type.9 Each individual can contribute to migrant

stock as both a rural-urban and international migrant.10

Migrants can also contribute to the

migrant stock of more than one household or village. For example, if a migrant moves to Dhaka

from Household Linage A, returns, moves to Household Lineage B, and moves on to Dhaka, he

would contribute to the migrant stock for each household lineage. A migrant cannot, however,

8 Theoretical expectations can be used to hypothesize both that return migrants should encourage migrants morethan current ones, because they have more social contact with prospective migrants, or less, because they are not in a

position to provide assistance. While these effects probably do not exactly cancel one another, they represent a topic

for further study. In either case, migration-out should be sufficient to identify important effect9 This seems the most straightforward choice given that this is an analysis of networks of people, not of moves.

There is no reason a priori to think that frequent moves increase the likelihood of chain migration.10 It is quite rare for an individual to have migrated to both destination types. Among men age 15-49 at the time of

the 1982 census, 24% would migrate to an urban area, 11% abroad, but only 2% to both.


19/47

17

contribute twice to the migrant stock of the same primary network. So, if a migrant moves twice

from separate households in the same household lineage, he would only be counted once.

Similarly, if household A and household B are in separate household lineages but in the same

bari, the migrant would be counted once for each primary network, but only once for the bari.

Step 6: Construction of Migrant History/Stock Measures

Finally, migrant stock measures are created for each of the eight reference groups ( i) discussed in

Steps 1-3: survey household, village, household lineage, bari, household lineage first- and

second-order peer networks, and bari first- and second-order peer network. For each reference

group i and destination typej (urban or overseas), the migrant stock ratio ( jim , ) is calculated

simply as the ratio of migrants ( jiM , ) divided by the total males age 15-49 present in the

reference group in the 1993 census ( jiT, )

ji

ji

jiT

Mm

,

,

, = .

For the substantial proportion of reference groups having no males at all (see Table 2 above), the

migrant stock ratio is set to the global mean migrant stock ratio for the entire study area.

Because the migration file does not distinguish between current and returned migrants, it

is not possible to identify a denominator for a standard migration prevalence rate. As such, the

migrant stock ratio, henceforth referred to as migrant stock, simply captures the ratio of

migrant males to males currently in the area.11

From a market standpoint, the migrant stock ratio

11 This use of non-standard denominator is common to ethnosurvey approaches to estimating migrant stock as well.

There are two possible methods for constructing standard denominators and prevalence measures using HDSS datathat could be applied in future studies. If current and former migrants are distinguished, then denominators can be

calculated directly as the sum of current migrants and current residents. Otherwise, the current population for each


20/47

18

is the most appropriate measure of migrant stock, however, since it reflects the relationship

between the total number of sources of migrant-specific social capital and the total number of

people who, as rural-resident males age 15-49, are potential migrants.

Table 3 summarizes migrant stock measures for each reference group. Because smaller

networks are more likely to have no migrants, the mean migrant stock ratio tends to increase for

higher-order networks. Medians are in all cases lower than means, indicating skewing to the

right. For the primary household network, less than 50% of respondents had any migrant stock.

Because the current survey household consists only of those not currently migrating, the migrant

stock estimate is substantially lower than for other reference groups. In general, the migrant

stock ratio for each reference group is about 0.25 for rural-urban migration (meaning that in a

particular reference group there exists 1 man who ever migrated for every 4 men currently living

in the study area) and 0.08 for international migration (1 ever-migrant for every 12 men in the

study area).


Table 4 addresses interrelationships among network migrant stock measures by

comparing zero-order correlations separately for rural-urban and international migrant stock. In

general, the measures appear to be unique. Correlations among bari networks tend to be higher

than for household networks because they are larger and more homogenous. At each level of

linkage (primary, 1st

order, 2nd

order), bari and original household migrant stocks tend to be

strongly correlated (for primary networks, the rural-urban migrant stocks are correlated at a 0.40

coefficient, while international stocks are 0.29). This is unsurprising given that household

networks are principally a subset ofbari networks. In this respect, it is useful to interpret

network can be identified from a combination of the 1982 census and subsequent death records under the

assumption that men did not move from one network to another.


21/47

19

coefficients for household network effects as their effect over and above the coefficients forbari-

level networks. Finally, because correlations between 1st- and 2

nd-orderbari networks are so high

(0.50 for international, 0.28 for rural-urban), 2nd

orderbari network measures are not employed

in the analysis.


Table 5 describes the interrelationships of network migrant stocks to contextual migrant

stock measures and to two key endowments that predict migration, household assets and the

respondents own schooling. In the first two columns, correlations with survey household and

village migrant stocks suggest that network migrant stock measures have strong external validity,

with all correlations running in a positive direction. At the same time, network migrant stock

measures provide unique information. Only 2nd

orderbari migrant stock is substantially similar

to the village level migrant stock (0.37 for rural-urban migration, 0.43 for international),

providing further justification for removing 2nd

orderbari networks from the analysis.

Unsurprisingly, bari migrant stocks are more strongly correlated to village migrant stocks, while

household lineage migrant stocks are more strongly correlated to survey household stocks. look

more similar to village migrant stocks than household.


The final two columns of Table 5, which correlate network migrant stocks to two SES

measures, offer further evidence of the external validity of network migrant stocks. As suggested

by the diffusion and migration framework, migrant stock measures are for the most part

positively correlated with household assets and particularly schooling. In other words, give that

past migrants tend to have higher SES and households tend to stratify socially according to SES,

it is unsurprising that those with high migrant stock also have high SES. The strength of this


22/47

20

relationship begins to break for higher order linkages and forbari networks versus household

networks, but the positive correlation to SES remains even for fairly distant connections.

These relationships offer further predictions for the statistical models. To some extent,

the effects of household assets and individual schooling on the hazard of migration should be

spurious. As migrant stock measures are added to a set of controls, the effects of assets and

schooling on migration should become more negative. Positive effects should be attenuated and

perhaps explained away, while negative effects should grow more negative.

Step 7: Matching study population to HDSS migration followup, 1997-2002

The final data management step needed to conducting an event history of analysis of migration

events involves linking the adult males in MHSS households to subsequent HDSS vital event

files. The privacy-protected version of MHSS identifies individuals according to their HDSS

Registration IDs (RIDs). MHSS was linked to HDSS files for the period from 1996 to 2002 as

part of an NIA-funded study of social networks and mortality. The sample began with the 5,831

adult males age 15 to 49 included in MHSS household rosters. Errors in the transcription of RIDs

in MHSS led to the exclusion of 227 men, 0r 3.9% of the study sample, who could not be

matched to the HDSS. This left 5,604 men. Complete household asset and schooling data were

not available for 254 men, or 4.5% of the remaining males.

This final study sample of males age 15-49 numbered 5,350. Of these individuals, 138, or

2.6%, died before the end of the followup period in December 2002; 946 individuals, or 17.7%

of the sample, made a rural-urban move; and 534, or 10.0%, moved abroad. Figure 1 graphs the

Kaplan-Meier cumulative failure distributions for rural-urban and international migration, with

adjustments for censoring due to rural-rural migration and death, and competing risks due to

migration to the other destination type. As suggested by the migrant stock ratios, rural-urban


23/47

21

migration is considerably more likely than international migration but both are quite common in

comparison to most other settings.

[INSERT FIGURE 1 ABOUT HERE]

VI. Statistical Methods and Results

The likelihood of rural-urban and international migration is tested using Cox Proportional

Hazard Models of survival. Separate models for each destination typejpredict the relative

change in the hazard (h)of migration by individualx at time tas the product of a baseline hazard

)(0, thi and a matrix of regressors measuring migrant stock ratios for the reference groups (i)

described above ( xjim ,, ) and individual and household control variables ( xc )

)exp()()(h ,,,0,xj, xxjijii cmtht +=

Models were adjusted for censoring due to death or rural-rural migration. Models of rural-urban

(international) migration were adjusted to account for the competing risks of international (rural-

urban) migration. Time to migration or censoring was estimated as the difference in years

between the date of first migration or death event and the date that the respondents MHSS

household roster was completed. HDSS provides a high level of specificity because the exact

date of migration or death was gathered from origin household members on a monthly basis.

Because all migrant stock measures for each reference group were highly skewed, models

explored both the raw stocks, logged stocks, and square and cubic terms. Only the best-fitting

models are presented. Is it ok to go back and forth between tenses like in the first and second

sentence? Except where noted, raw and logged stocks had similar levels of association with

subsequent migration. Models do not account for migrant stock to the alternate destination type.


24/47

22

In other words, models of rural-urban (international migration) do not account for past

experience of international (rural-urban) migration.

Models include controls for individual and household demographic and socioeconomic

characteristics at baseline. For the individual, these include age (expressed in five-year age

categories), marital status (married vs. not married), completed schooling (divided into five

groups = 0, 2-4, 5-9, 10-11, 12+); and prior years of migration experience. The models allow all

age effects to vary over time. Models also control for household income and assets at baseline.

The addition of migrant stock measures should improve estimates of socio-demographic

controls. For instance, since younger people have little experience as migrants, estimating

migrant stock should lower the age coefficient. Since wealthier households and better-educated

individuals tend to have stronger social networks, migrant stock controls should lower asset and

education coefficients.

Extensive robustness testing was conducted for each statistical model; this includes tests

of the proportional hazards assumption by testing the trend of Schoenfield residuals for each

model specification. This assumption was satisfied in most cases, but not all. Some specifications

suggested that the proportionality assumption was violated for some but not all categories of the

educational measure. Time-varying covariates for education are not included in order to ensure

that the education coefficients are transparent for the reader to interpret. All specifications were

estimated with and without time-varying education covariates, and all key findings with respect

to the effects of migrant stock were robust to such changes. Each specification was also

replicated using discrete-time logistic hazard regression models with time-varying interaction

terms for age and education. The effects of all migrant stock measures were again robust to this

change. The proportional hazard models were selected for their interpretability and, also, to take


25/47

23

advantage of the high level of specificity in measuring the timing of migration events. Using the

method identified by May and Homser (1998), goodness-of-fit tests compared the observed and

expected number of migration events for the total sample and for each decile of risk.

Presentation of results will address models of rural-urban and international migration

separately. With the addition of each primary and peer network migrant stock measures, I will

revisit the strength of the village-level contextual migrant stock effect and summarize these

changes at the end. Fixed-effects models will also be employed to better isolate network effects.

Finally, I will separate peer network migrant stock effects into migrant stock for peers living in

the same village and in different villages.

Models of Rural-Urban Migration

Table 5 presents a series of proportional hazard models predicting the hazard of rural-urban

migration for the study population in terms of rural-urban migrant stock measures. The base

model 1 includes the socio-demographic control variables and migrant stock measured at the

village level. According to the May-Homser goodness-of-fit test, this model predicted 98% of

observed events. Model predictions, however, predicted significantly fewer events than were

actually observed for respondents in the seventh decile of risk, and significantly more events for

the eighth decile.

Covariates for village-level migrant stock and the socio-demographic control variables

corresponded to those observed in most other settings. The hazard of rural-urban migration was

higher among those who were younger and unmarried, those from households with fewer assets,

and those with greater schooling.12

The respondents own migration experience had a positive

12 Effects for age were estimated in five-year ago categories, including both main effects and time-varying effects. In

the interest of simplicity, age effects are not shown in Tables 6 and 8 (for international migration) in the text.


26/47

24

association with the hazard of further migration. Current household income was not found to

have a significant effect, and was dropped from all models.


Village-level migrant stock was found to be a highly significant predictor of migration; a

one standard deviation increase (by 0.11) was associated with a 15% increase in the hazard of

migration. It is important to note, however, that the distribution of village-level migrant stock is

highly skewed to the right. The much smaller difference in village-level migrant stock between

the 50th

and 75th

percentile is only 0.06, which would result in an 8% increase in the hazard of

migration. Migrant stock in the survey household, however, was not found to be significantly

associated with the hazard of migration, and therefore was not included in the model.

The village migrant stock effect estimated in Model 1 establishes a baseline for

comparing to each of the following models, which incrementally introduce primary and peer

network effects. Model 2 introduces a covariate for the migrant stock ratio among members of

the household lineage, which has a positive association with the hazard of migration that is

significant at the p


27/47

25

effect was of a much lower magnitude than household lineage and village effects. A one standard

deviation change in bari migrant stock (0.29) results in only a 4% increase in the hazard of

migration. A shift from the 50th

to 75th

percentile values (0.19 to 0.33) would result in only a 2%

change. As mentioned above, because the household lineage is almost a perfect subset of the

bari, the diminished but still highly significant coefficient for household lineage migrant can

now be interpreted as its effect over and above the bari effect. A man from a household lineage

whose migrant stock was 0.33 higher than his barismigrant stock would have a 9% greater

hazard of migration than the average household in his bari. The migrant stock of both patrilineal

kin groups is important, but household lineage stock is substantially more important. The village-

level effect was further reduced.

Model 4 and 5 introduced the effects of first- and second-order peer networks linked

through the household lineage. Model 4 included migrant stock measures separately for first and

second order networks; each effect was positive, but the coefficients fell below the 5%

significance level. Given the limited size of the first order network, and the high proportion of

respondents having no males in their household peer networks, Model 5 tested the effect of the

migrant stock for the two household lineage peer networks combined. This measure had a

positive association with the hazard of migration, significant at the p


28/47

26

Model 6 introduced the effect of migrant stock in the first-orderbaripeer network, which

again has a positive and highly significant association (p


29/47

27

strength of social ties within a village, or do they also reflect the influence of peers living outside

the village? The tests shown in Table 7 offer fairly simplistic attempts to address this question by

separating the peer migrant stock ratio effects into ratios for peers living inside and outside the

village. In both the case of household and baripeers, the models show no statistically significant

differences between the effect of those inside or outside the village. We therefore can neither say

that peers from outside or inside the the village are more important. In other words, differences

in peer networks leading to a greater hazard of rural-urban migration for some households reflect

both differences in the strength of ties within and outside the community.


Models of International Migration

Table 8 presents proportional hazard models of international migration. Model 1 first included

socio-demographic controls and contextual migrant stock effects. Goodness-of-fit tests show that

the expected number of events equals 101% of the total observed events, and expected results

only differ significantly from observed in the fourth decile of risk.


As with the rural-urban migration models, socio-demographic effects correspond to those

found in previous studies. Unlike rural-urban migration, international migration is more likely

among respondents from households with higher assets. This reflects the high costs and risks of

international migration, and should be diminished by the inclusion of migrant stock controls. Net

of the asset effect, households with higher incomes are somewhat less likely to move abroad.

Overall, schooling has a positive association with international migration. Schooling effects,

however, are much smaller than for rural-urban migration, and only respondents with 5-9 or 10-


30/47


31/47

29

Models 4, 5, and 6 introduce household lineage and baripeer network effects. While first

order household peers are not statistically associated with the hazard of migration, second order

household peer effects are moderately significant (at the p


32/47

30

outside the village have as much explanatory power as those inside the village. A strictly

contextual model could not capture these effects, overlooking the critical importance of extra-

community ties in allowing migration opportunities to jump from community to community.

The results also highlight a tendency for contextual models of migration to produce

spuriously positive estimates for the sociodemographic correlates of migration. In all cases,

Table 5 found a positive correlation between lineage and peer network migrant stock and

measures of physical and human capital. Village migrant stocks, on the other hand, are largely

uncorrelated with endowments. As a result, models without network effects overstated the

positive effects of schooling, overstated the positive effects of assets on international migration,

and understated the negative association between assets and rural-urban migration.

It is important to be critical when integrating existing models of migration at the

individual, household, and community levels (Arango 2000). The typical individual and

household factors entered into migration models (education, wealth, income, age) tend to specify

a relationship between stratification and migration. Yet community-based measures of migrant

stock inherently restrict the array of possible results to those that imply declining within-

community inequality. While the contextual approach may effectively differentiate between

communities with better or worse overall endowments, it cannot address persistent within-

community heterogeneity in individual and household access to community endowments such as

migration-specific social capital. The multidimensional process of chain migration reflects both

contextual and contingent opportunities.

Three major questions remain in addressing the extent to which the chain migration

process is contextual or conditional. First, are contextual and conditional forms of migration-

specific social capital complements or substitutes? In other words, is it sufficient to have either


33/47

31

peer-based or community-based access to migration, or is it necessary to have both? A test of the

interactions between contextual and network migrant stock measures can answer this question.

Second, are the various forms of migration-specific social capital complements or

substitutes to individual and household endowments such as education and assets? If they are

substitutes, then migration-specific social capital could narrow socioeconomic disparities in

access to migration. If they are complementary, however, we would identify another source of

exclusion in the chain migration process. Even households that have high strong peer networks

may only be able to utilize them through the direct or indirect expenditure of resources as a

buy-in to the network (Portes and Sensenbrenner 1993; Portes 1998; Stark and Wong 2002).

This concern can be addressed by testing interactions between migrant stock and endowments.

Finally, it remains important to understand variations in the quality of migration

opportunities, and how the quality of migration opportunities might diminish as the chain

migration process unfolds. Do households that move first and bear substantial risk benefit more

from migration than those who move later, expending little human or social capital?

This study has begun to address the functional mechanisms underlying chain migration,

and has set the stage for understanding the implications of chain migration for household

migration opportunities. Although the migration process can affect non-migrant households in a

community through a variety of indirect pathways, our understanding of the relationship between

migration and within-community opportunity and inequality ultimately depends on the pathways

of transmission of migration opportunities themselves (Taylor 1992; Taylor and Wyatt 1996).

Until we better understand how chain migration, migrant networks, and migration-specific social

capital operate, it would be safer to assume that their effects on inequality are neutral, not

positive.


34/47

32

References

Aguilera, Michael B. 2003. "The Impact of the Worker: How Social Capital and Human CapitalInfluence the Job Tenure of Mexican Immigrants." Sociological Inquiry 73(1):52-83.

Aguilera, Michael B., and Douglas S. Massey. 2003. Social Capital and the Wages of MexicanMigrants: New Hypotheses and Tests. Social Forces 82(2): 671-701.

Arango, Joaquin. 2000. Explaining Migration: A Critical View.International Social ScienceJournal52(3): 283-296.

Burt, Ronald S. 1998. The Gender of Social Capital. Rationality and Society 10(1): 5-46.

Burt, Ronald S. 2000. The Network Structure of Social Capital.Research in OrganizationalBeahvior22: 345-423.

Carrington, William J., Enrica Detragiache, and Tara Vishwanath. 1996. Migration with

Endogenous Moving Costs.American Economic Review 86(4): 909-930.

Collyer, Michael. 2005. When do Social Networks Fail to Explain Migration? Accounting for

the Movement of Algerian Asylum-Seekers to the UK.Journal of Ethnic and MigrationStudies 31(4): 699-718.

Curran, Sara R., and Estela Rivero-Fuentes. 2003. Engendering Migrant Networks: The Case of

Mexican Migration.Demography 40(2): 289-307.

Das, Raju J. 2004. Social Capital and Poverty of the Wage-Labour Class: Problems with the

Social Capital Theory. Transactions of the Institute of British Geographers 29(1): 27-45.

Davis, Benjamin, and Paul Winters. 2001. Gender, Networks and Mexico-US Migration.Journal of Development Studies 38(2): 1-26.

Davis, Benjamin, Guy Stecklov, and Paul Winters. 2002. Domestic and International Migration

from Rural Mexico: Disaggregating the Effects of Network Structure and Composition.

Population Studies 56(3): 291-309.

Durand, Jorge, Douglas S. Massey, Rene M. Zenteno. 2001. Mexican Immigration to the UnitedStates: Continuities and Changes.Latin American Research Review 36(1): 107-127.

Ebaugh, Helen Rose and Mary Curry. 2000. Fictive Kin as Social Capital in New Immigrant

Communities. Sociological Perspectives 43(2): 189-209.

Elmhirst, Rebecca. 2002. Daughters and Displacement: Migration Dynamics in an Indonesian

Transmigration Area.Journal of Development Studies 38(5): 143-166.

Espinosa, Kristin E., and Douglas S. Massey 1997. "Undocumented Migration and the Quantity

and Quality of Social Capital." Soziale Welt12:141-62.

Foster, Andrew D. 2005. Altruism, Household Coresidence and Womens Health Investment in

Rural Bangladesh. Unpublished Manuscript, Brown University


35/47

33

Greenwell, Lisa, R. Burciaga Valdez, and Julie DaVanzo. 1997. Social Ties, Wages, and

Gender in a Study of Salvadorean and Filipino Immigrants in Los Angeles. Social ScienceQuarterly 78(2): 559-577.

Hagan, Jacqueline. and Helen Rose Ebaugh. 2003. Calling upon the Sacred: Migrants' Use of

Religion in the Migration Process.International Migration Review 37(4): 1145-1162.

Joshi Shareen and Nistha Sinha. 2005. The Effect of Household Partition on Childrens Workand Schooling in Rural Bangladesh. Unpublished Manuscript, Yale University.

Kandel, William, and Douglas S. Massey. 2002. The Culture of Mexican Migration: A

Theoretical and Empirical Analysis. Social Forces 80(3): 981-1004.

Krishna, Anirudh. 2004. Understanding, Measuring and Utilizing Social Capital: ClarifyingConcepts and Presenting a Field Application from India.Agricultural Systems 82(3): 291-305.

Krissman, Fred 2005. Sin Coyote Ni Patron: Why the Migrant Network Fails to Explain

International Migration.International Migration Review 39(1): 4-44.

Kuhn, Randall. 2003. Identities in Motion: Social Exchange Networks and Rural-Urban

Migration in Bangladesh. Contributions to Indian Sociology 37: 311-337.

Kuhn, Randall. 2005. The Effects of Fathers and Siblings Migration on Childrens Pace ofSchooling in a Migrant-Sending Region of Bangladesh. Workshop on The Impacts of

Migration on the "Left-Behind" in Asia, Asian MetaCentre, National University of

Singapore, Hanoi, Vietnam, March 2005.

Lindstrom, David P., and Douglas S. Massey. 1994. Selective Emigration, Cohort Quality, and

Models of Immigrant Assimilation. Social Science Research 23:315-349.

Lindstrom, David P., and Nathanael Lauster. 2001. Local Economic Opportunity and theCompeting Risks of Internal and US Migration in Zacatecas, Mexico.InternationalMigration Review (354): 1232-1256.

Massey, Douglas S. 1990. Social Structure, Household Strategies, and the Cumulative

Causation of Migration.Population Index 56:3-26.

Massey, Douglas S., Luin P. Goldring, and Jorge Durand. 1994. "Continuities in Transnational

Migration: An Analysis of 19 Mexican Communities."American Journal of Sociology99:1492533.

Massey, Douglas S., Rafael Alarcn, Jorge Durand, and Humberto Gonzlez. 1987.Return toAztlan: The Social Process of International Migration from Western Mexico. London,England: University of California Press.

Massey, Douglas S., Joaqun Arango, Graeme Hugo, Ali Kouaouci, Adela Pellegrino, and J.Edward Taylor. 1999. Worlds in Motion: Understanding International Migration at The Endof The Millennium. New York: Oxford University Press.


36/47

34

May, S., and D.W. Hosmer. 1998. A Simplified Method of Calculating an Overall Goodness-of-

Fit Test for the Cox Proportional Hazards Model.Lifetime Data Analysis, 4: 109-120.

Menjivar, Cecilia. 2002. The Ties That Heal: Guatemalan Immigrant Women's Networks and

Medical Treatment.International Migration Review 36(2): 437-466.

Munshi, Kaivan. 2003. Networks in the Modern Economy: Mexican Migrants in the US LaborMarket. Quarterly Journal of Economics 118(2): 549-599.

Palloni, Alberto, Douglas S. Massey, Miguel Ceballos, Kristin Espinosa, and Michael Spittel.

2001. Social Capital and International Migration: A Test Using Information on Family

Networks.American Journal of Sociology 106(5): 1262-1298.

Portes, Alejandro. 1998. Social Capital: Its Origins and Applications in Modern Sociology.

Annual Review of Sociology 24:1-24.

Portes, Alejandro. 2000. The Two Meanings of Social Capital. Sociological Forum 15(1):1-12.

Portes, Alejandro and Robert L. Bach. 1985.Latin Journey: Cuban and Mexican Immigrants inThe United States. Berkeley: University of California Press.

Rosenzweig, Mark R., and Oded Stark. 1989. Consumption Smoothing, Migration, and

Marriage.Journal of Political Economy 97: 905-926

Sanders, Jimy, Victor Nee, Scott Sernau. 2002. Asian Immigrants' Reliance on Social Ties in aMultiethnic Labor Market. Social Forces 81(1): 281-314.

Shah, Nasra M. 2000. Relative Success of Male Workers in the Host Country, Kuwait: Does theChannel of Migration Matter?International Migration Review 34(1): 59-78.

Shah, Nasra M., and Indu Menon. 1999. Chain Migration through the Social Network:Experience of Labour Migrants in Kuwait. International Migration 37(2): 361-382.

Silvey, Rachel, and Rebecca Elmhirst. 2003. Engendering Social Capital: Women Workers andRural-Urban Networks in Indonesia's Crisis. World Development31(5): 865-879.

Stark, Oded, and You Qiang Wang. 2002. Migration dynamics.Economics Letters 76(2): 159-164.

Strang, David, and Sarah A. Soule. 1998. Diffusion in Organizations and Social Movements:

From Hybrid Corn to Poison Pills.Annual Review of Sociology 24: 265-290.

Taylor, J. Edward. 1986. Differential Migration, Networks, Information, and Risk. Pp. 147-171 inResearch in Human Capital and Development: Migration, Human Capital, andDevelopment, Volume 4, edited by Oded Stark. Greenwich, CT: JAI Press, Inc.

Taylor, J. Edward. 1992. Remittances and Inequality Reconsidered: Direct, Indirect, andIntertemporal Effects.Journal of Policy Modeling14:187-208.


37/47

35

Taylor, J. Edward and T. J. Wyatt. 1996. The Shadow Value of Migrant Remittances, Income

and Inequality in a Household-farm Economy. The Journal of Development Studies32(6):899-912.

Van der Gaag, Nicole, and Leo Van Wissen. 2002. Modelling Regional Immigration: Using

Stocks to Predict Flows.European Journal of Population 18(4): 387-409.

Waldorf, Brigitte. 1996. The Internal Dynamic of International Migration Systems.

Environment and Planning A 28(4): 631-650.

Wegge, Simon A. 1998. Chain Migration and Information Networks: Evidence from

Nineteenth-Century Hesse-Cassel.Journal of Economic History 58(4): 957-986.

Wejnert, Barbara. 2002. Integrating Models of Diffusion of Innovations: A ConceptualFramework.Annual Review of Sociology 28: 297-326.

Wilson, Tamar Diana. 1998. Weak Ties, Strong Ties: Network Principles in Mexican

Migration.Human Organization 57(4): 394-403.

Winters, Paul, Benjamin Davis, and Leonardo Corral. 2002. Assets, Activities and Income

Generation in Rural Mexico: Factoring in Social and Public Capital.AgriculturalEconomics 27(2): 139-156.

Winters, Paul, Alain de Janvry, and Elizabeth Sadoulet. 2001. Family and CommunityNetworks in Mexico-US Nigration.Journal of Human Resources 36(1): 159-184.

Zhao, Yaohui. 2003. The Role of Mmigrant Networks in Labor Migration: The Case of China.

Contemporary Economic Policy 21(4): 500-511.


38/47

36

Table 1: Migrant Network Levels of Aggregation Summarized

Traditional Primary Networks

Survey Household

Village

New Primary Networks New Peer NetworksHousehold lineage 1st order household peers 2nd order household peers

Bari(compound) lineage 1st orderbaripeers 2nd orderbaripeers


39/47

37

Table 2: Summary of Reference Group Composition

Reference Group Links Males

Mean S.D. Max Mean S.D. Max

Contextual units

MHSS household 2.4 1.2 8

Village 1351.8 1067.4 4181

Household lineage networks

Primary network 5.3 6.1 66

First-order peers 1.7 1.7 13 7.4 11.5 123

Second-order peers 3.2 4.5 43 13.4 23.3 302

Barinetworks

Primary network 29.2 22.6 226

First-order peers 11.8 9.6 55 335.1 312.6 1900

Second-order peers 152.4 141.3 850 4384.9 4366.5 23304

Source: HDSS (1974-1996)


40/47

38

Table 3: Summary of Migrant Stock for each Reference Group

Reference Group Urban Migration International Migration

Mean SD Median Mean SD Median

Contextual units

MHSS household 0.074 0.147 0.000 0.018 0.075 0.000

Village 0.252 0.107 0.241 0.084 0.061 0.066

Household lineage networks

Primary network 0.219 0.421 0.000 0.058 0.185 0.000

First-order peers 0.279 0.399 0.245 0.093 0.199 0.083

Second-order peers 0.286 0.439 0.245 0.086 0.153 0.083

Bari

Primary network 0.255 0.291 0.188 0.084 0.173 0.042

1st

order peers 0.257 0.277 0.237 0.079 0.097 0.063

2nd

order peers 0.237 0.075 0.243 0.077 0.058 0.065



41/47

39

Table 4: Zero Order Correlations Among Network Migrant Stock Ratios

Panel A: Rural-Urban Migration

Household Lineage Bari Lineage

PrimaryNetworks

Peer NetworksPrimary

NetworksPeer Networks

1st

order 2nd

order 1st

order 2nd

order

Household Lineage

Primary 1.00

1st Order Peer 0.05 1.00

2nd Order Peer 0.01 0.07 1.00

Bari Lineage

Primary 0.40 0.10 0.03 1.00

1st Order Peer 0.05 0.14 0.01 0.08 1.00

2nd Order Peer 0.12 0.12 0.09 0.17 0.28 1.00

Panel B: International Migration

Household Lineage Bari Lineage

PrimaryNetworks

Peer NetworksPrimary

NetworksPeer Networks

1st

order 2nd

order 1st

order 2nd

order

Household Lineage

Primary 1.00

1st Order Peer 0.06 1.00

2nd Order Peer 0.01 0.09 1.00

Bari Lineage

Primary 0.29 0.05 0.04 1.00

1st Order Peer 0.05 0.15 0.11 0.22 1.002nd Order Peer 0.03 0.10 0.13 0.31 0.50 1.00



42/47

40

Table 5: Zero Order Correlation of Peer Network Migrant Stock Ratios toConventional Migrant Stock Ratios and SES Measures

Rural-Urban Migration

Migrant Stock Ratio SES

Household Lineage Survey HH Village HH Assets Schooling

Primary 0.32 0.10 0.06 0.10

1st Order Peer 0.00 0.11 0.06 0.09

2nd Order Peer 0.00 0.04 0.05 0.06

Bari Lineage Survey HH Village HH Assets Schooling

Primary 0.18 0.24 0.05 0.09

1st Order Peer 0.04 0.19 0.05 0.07

2nd Order Peer 0.09 0.37 0.14 0.13

International Migration

Migrant Stock Ratio SES

Household Lineage Survey HH Village HH Assets Schooling

Primary 0.37 0.07 0.12 0.13

1st Order Peer 0.05 0.07 0.11 0.10

2nd Order Peer -0.00 0.14 0.08 0.05

Bari Lineage Survey HH Village HH Assets Schooling

Primary 0.12 0.34 0.04 0.05

1st Order Peer 0.01 0.33 0.01 0.04

2nd Order Peer -0.00 0.43 -0.00 0.01

Source: MHSS (1996) and HDSS (1974-1996)


43/47

41

Table 6: Covariates from Cox Proportional Hazard Model of Rural-Urban Migration

Demographic / SocioeconomicControls (1) (2) (3) (4) (5) (6)

-0.404* -0.395* -0.372* -0.368* -0.364* -0.362*Respondent Married(0.176) (0.176) (0.176) (0.176) (0.177) (0.176)

-0.104** -0.109** -0.111** -0.116** -0.118** -0.120**Household assets (logged)

(0.036) (0.037) (0.037) (0.037) (0.036) (0.036)

Schooling (0 years = reference)

0.514** 0.510** 0.501** 0.501** 0.504** 0.502**2-4 years

(0.167) (0.165) (0.164) (0.165) (0.164) (0.165)

0.421** 0.397** 0.368* 0.360* 0.357* 0.353*5-9 years

(0.153) (0.153) (0.155) (0.153) (0.154) (0.154)

0.858** 0.824** 0.799** 0.792** 0.782** 0.774**10-11 years

(0.172) (0.172) (0.176) (0.176) (0.176) (0.177)

1.071** 1.027** 0.998** 0.992** 0.996** 0.992**12+ years

(0.239) (0.235) (0.234) (0.233) (0.235) (0.236)0.068** 0.052* 0.048 0.048 0.047 0.047Total Years Living in City

(0.023) (0.026) (0.026) (0.025) (0.025) (0.025)

Migrant Stock Ratios (1) (2) (3) (4) (5) (6)

1.252** 1.189** 1.004** 0.963** 0.939** 0.868**Village level

(0.288) (0.292) (0.298) (0.298) (0.300) (0.304)

Primary network level

0.373** 0.258** 0.258** 0.254** 0.252**Household lineage

(0.081) (0.096) (0.095) (0.096) (0.096)

0.126** 0.121** 0.121** 0.120**Bari(logged)

(0.041) (0.041) (0.041) (0.041)

Peer network level

0.165Household 1st order(0.090)

Household 2nd

order 0.079

(0.060)

Household 1st/2

norder 0.353** 0.343**

(0.131) (0.132)

0.182**Bari1st

order

(0.055)

Sample Diagnostics (1) (2) (3) (4) (5) (6)

F-Test Statistic: 474.65 487.54 500.7 514.13 534 634.52

Degress of Freedom 20 21 22 24 23 24

???? 0.03 0.04 0.04 0.04 0.04 0.06

Source: MHSS (1996) and HDSS(1974-2002)Sample size for all models = 5,350Robust standard errors in parentheses; * significant at 5%; ** significant at 1%


44/47

42

Table 7: Separating Peer Network Effects on Rural-Urban Migration into SameVillage and Different Village

Household 1st & 2nd order peers (1) (2) (3) (4)

0.343** 0.342**Combined

(0.132) (0.132)

0.230** 0.229**Same Village

(0.085) (0.086)

Different Village 0.093 0.088

(0.115) (0.115)

Chi-Square Difference Test 0.85 0.89

Bari1st order peers (1) (2) (3) (4)

0.182** 0.196** Combined

(0.055) (0.051) Same Village 0.056 0.039

(0.109) (0.116)

Different Village 0.169** 0.186***

(0.056) (0.049)

Chi-Square Difference Test 0.77 1.25

Source: MHSS (1996) and HDSS(1974-2002)Sample size for all models = 5,350Robust standard errors in parentheses; * significant at 5%; ** significant at 1%


45/47

43

Table 8: Covariates from Cox Proportional Hazard Model of International Migration

Demographic/Socioeconomic Controls (1) (2) (3) (4) (5) (6)

0.227** 0.208** 0.197** 0.193** 0.190** 0.190**Household assets (logged)

(0.054) (0.053) (0.053) (0.053) (0.054) (0.053)

Household income (logged) -0.034* -0.031* -0.030 -0.030 -0.029 -0.030

(0.015) (0.015) (0.016) (0.016) (0.016) (0.016)0.008 -0.002 0.001 0.002 0.001 -0.0012-4 years schooling

(0.242) (0.233) (0.235) (0.236) (0.235) (0.235)

0.561** 0.539** 0.530** 0.525** 0.522** 0.529**5-9 years schooling

(0.163) (0.163) (0.162) (0.163) (0.162) (0.163)

0.384* 0.313 0.298 0.309 0.301 0.306Completed lower secondary

(0.188) (0.188) (0.185) (0.185) (0.183) (0.182)

0.142 0.167 0.159 0.179 0.176 0.194Completed upper secondary or higher

(0.374) (0.365) (0.364) (0.365) (0.364) (0.365)

0.165** 0.124* 0.119 0.119 0.119 0.120*Respondent's Past Migration Years

(0.057) (0.060) (0.061) (0.061) (0.061) (0.061)

Migrant Stock Ratios (1) (2) (3) (4) (5) (6)

18.7** 17.4** 14.9** 13.9* 13.7* 13.6**Village migrant stock ratio

(6.28) (6.11) (6.13) (6.12) (6.16) (6.2)

-93.9** -89.3** -80.8* -74.7* -74.2* -70.8*Village migrant stock ratio (squared)

(33.0) (32.1) (31.7) (31.7) (31.7) (32.8)

Village migrant stock ratio (cubed) 99.6** 95.2** 86.4* 74.1* 73.4* 68.8

(37.1) (36.2) (35.7) (35.3) (35.4) (36.8)

Primary Extended Networks

0.996** 0.780** 0.792** 0.793** 0.797**Household Lineage

(0.149) (0.231) (0.227) (0.227) (0.230)

0.133* 0.136* 0.134* 0.138*Bari(logged)

(0.054) (0.054) (0.054) (0.054)

Peer Networks-0.192Household 1

storder

(0.201)

0.673** 0.652* 0.680**Household 2nd

order

(0.251) (0.255) (0.246)

Bari1st

order -0.936

(1.220)

Sample Diagnostics (1) (2) (3) (4) (5) (6)

F-Test Statistic: 306.84 401.65 453.43 475.19 469.12 466.53

Degrees of Freedom 22 23 24 26 25 26

Source: MHSS (1996) and HDSS(1974-2002)Sample size for all models = 5,288

Robust standard errors in parentheses; * significant at 5%; ** significant at 1%


46/47

44

Figure 1: Kaplan-Meier Estimates of Cumulative Failure Distribution for Rural-Urban and


0

0.05

0.1

0.15

0.2

0 1 2 3 4 5 6 7 8

Years

CumulativeFailureRate

Rural-Urban Migration



47/47

Figure 2: Estimated Impact of Village Migrant Stock on Relative Hazard of MigrationWith and Without Controlling for Network Effects, by Migrant Destination Type

Urban-Rural Migration

0.9

1.1

1.3

1.5

1.7

0 0.1 0.2 0.3 0.4

Village Migrant Stock

RelativeHazardofMigration


0.5

1

1.5

2

0 0.1 0.2 0.3

Village Migrant Stock

RelativeHazardofMigration

Notes: Lower line indicates model with network effects, upper line without network effects.

community or connections? a social networks approach to chain migration

Documents